Drug-Target Binding Affinity Prediction by Combination Graph Neural Network and BiLSTM

Drug development is not only expensive and time-consuming, but often companied with safety problems. The existing methods for predicting the affinity between drugs and targets are plagued with accuracy issues or have weak molecular feature capturing abilities. In this study, we propose a novel method that Combination Graph Neural Network (CGNN) and Bidirectional Long Short-Term Memory (BiLSTM) to predict drug and target affinity (DTA). Our approach employs the CGNN module to capture molecular features of drug compounds and utilizes the BiLSTM layer to process protein sequences. The drug and protein representations are then concatenated for the final prediction. Our model was trained on the Davis and KIBA (Kinase Inhibitor Bioactivity) benchmark datasets, and the experimental results demonstrate that it outperforms existing deep learning methods in DTA prediction and has excellent feature capturing ability.